The Ubuntu Linux Performance Over The Past Six Years On An Intel Xeon Server

[phoronix]

Phoronix: The Ubuntu Linux Performance Over The Past Six Years On An Intel Xeon Server

In needing to make some room in the racks for some new hardware and some other interesting platforms on the way, I've retired the last of the Intel Nehalem era hardware at Phoronix that was still used for occasional historical Linux performance tests... I decided to take this Sun Microsystems SunFire X4170 server with dual Intel Xeon E5540 (Nehalem EP) processors for a final spin before pulling it from the racks. Here is a look at how the near-final Ubuntu 18.10 Linux performance compares to that of Ubuntu 12.10.

http://www.phoronix.com/vr.php?view=26930

[schmidtbag]

I didn't know the mitigations were applied to hardware this old. Sure seems to have some pretty serious blows in performance though. But, I suppose stuff like this will help encourage people to move onto new hardware.

[torsionbar28]

I didn't know the mitigations were applied to hardware this old.

Intel and the server vendors are not producing new firmware/microcode for older vulnerable hardware. But that doesn't prevent the OS from applying software-based mitigations, which I believe is the case here.

Sure seems to have some pretty serious blows in performance though. But, I suppose stuff like this will help encourage people to move onto new hardware.

Looks like massive 25%+ hits in certain applications. I think a lot of new servers in the world will be sporting AMD EPYC chips after this spectre/meltdown debacle.

[skeevy420]

Intel and the server vendors are not producing new firmware/microcode for older vulnerable hardware. But that doesn't prevent the OS from applying software-based mitigations, which I believe is the case here.


Looks like massive 25%+ hits in certain applications. I think a lot of new servers in the world will be sporting AMD EPYC chips after this spectre/meltdown debacle.

My Dell T5500 x5687 Westmere box recently got a new firmware for spectre/meltdown. AFAIK, the T5500 is as far back as Dell is going in their firmware updates...so I got lucky with my old hardware.

All these results do is encourage me to make a Funtoo based system with InBetweenName's Gentoo 03-Graphite-LTO overlay since I can't afford new hardware.

[Niarbeht]

I hate to get all meme-y here, but...

Press F to pay respects.

[ms178]

My Dell T5500 x5687 Westmere box recently got a new firmware for spectre/meltdown. AFAIK, the T5500 is as far back as Dell is going in their firmware updates...so I got lucky with my old hardware.

All these results do is encourage me to make a Funtoo based system with InBetweenName's Gentoo 03-Graphite-LTO overlay since I can't afford new hardware.

I also have a Westmere and Sandy Bridge System. Unfortunately my Linux knowledge is not profound enough for this level of optimization yet. ;( But at least it is enough to compile my Kernel with these aggressive compiler flags and a custom config. The savings in RAM usage and improved snappiness are great! I'd love to see distros pushing a lot harder on this front though. The old

mantra

"it provides only 5% perf improvements at most and breaks the world" does not survive my own anecdotal evidence these days. As you say, this Funtoo overlay really pushes the limits and I very much concur with their reasoning in doing so. Why couldn't other distros have both: a generic repository as is the norm today and a fully O3-Graphite-LTO-march-native-optimized repo for recent / popular architectures? That could be even automated during the installation process. It would be surely a lot of work to maintain, but they could kill a lot of compiler and build system bugs upstream on their way and would be a meaningful innovation in the open source world. I also hope the Kernel patches for LTO support get mainlined soon. But unfortunately all these security mitigations were more of a focus of Kernel developers during this year.

[Enverex]

I didn't know the mitigations were applied to hardware this old. Sure seems to have some pretty serious blows in performance though. But, I suppose stuff like this will help encourage people to move onto new hardware.

It actually applies to everything all the way back to the original Pentium in some way, shape or form. The CPUs affected are all but the most recent generation that have a few features built-in to mitigate some of the new issues in hardware.

[blueweb]

Very interesting article. I must say that the slower performance on the newer OS was disappointing... I was hoping for the opposite.

Nice to see R-benchmark and pybench results have improved with time, but how much of that is due to improvements in R / python vs the OS?

[brauliobo]

conclusion: linux is getting slower!?!